Articles in press

Previous issues

¿Why residuals are symmetric in genetic evaluation?

Download paper Ver Versión Española

Authors: L. Varona, N. Ibáñez, R.N. Pena, J.L. Noguera, R. Quintanilla
Issue: 102-2 (98-102)
Topic: Animal Production
Keywords: Genetic evaluation, bayesian inference, asymmetric distributions, prolificacy

Statistical models for genetic evaluation make use of Gaussian or t residual distributions. However, some new statistical development allows using asymmetric distribution for the residual not controlled by the model. We have analysed a data set of litter size on pigs consisting of 2072 data for number of piglets born alive from 657 sows, with a total pedigree of 765 individual-sire-dam triplets. The model includes order of parity and herd-year-season systematic effects, Gaussian additive genetic effects and Gaussian permanent environmental effects. In addition, we use three different distributions for the residuals. 1) Symmetric Gaussian distribution, 2) Asymmetric Gaussian distribution and 3) Asymmetric t distribution. The three statistical models were compared using a Bayes Factor. The most suitable model corresponds to the asymmetric Gaussian distribution (Model 2). The posterior mean of heritability was 0.063, with a posterior standard deviation of 0.028. The resulting distributions of the residuals are strongly asymmetric, indicating that the sources of variation not controlled by the model have mostly a negative influence on the prolificacy. The asymmetry parameter can be understood as a measure of sensibility to negative environmental influences on the phenotype.

Back to Issue 102-2


 -   -   -   -   -   -   -